Generally, you should get all the data you can that is reasonably likely to relate to what people are interested in.
As far as integrating different kinds of information, see my talk from Buzzwords last yet on multi-modal recommendation. http://www.slideshare.net/tdunning/buzz-wordsdunningmultimodalrecommendation https://www.youtube.com/watch?v=fWR1T2pY08Y On Wed, Jan 15, 2014 at 12:27 AM, abhishek kumar < [email protected]> wrote: > Thanks Ted, it was very helpful.Though I still have some doubts: > > The click log would contain URL's and engagement of user as you suggested, > how can we use this data for building a recommend-er engine in mahout ? > > I started working with topic modelling,(I know mahout has LDA ) can it be > used for recommendation ? > > I only want to recommend web pages rather than making product based > recommender system, do I need to record user engagement for that also? > > > On Wed, Jan 15, 2014 at 4:06 AM, Ted Dunning <[email protected]> > wrote: > > > Manuel is close to what I meant. > > > > Just taking click-logs is a good start, but you also have to look at the > > user experience and decide what actions that people take that indicates > > engagement. > > > > For example, if you look at amazon, I would say that if you scroll down > the > > page to see the related items and then scroll down to see the reviews, > > these are both signs of engagement with a product. This engagement is > > distinctly better than just loading a page or clicking on a product link. > > > > Neither of these actions would appear in a click log since neither > involves > > a click. It is quite plausible that these actions would have different > > weight for recommendations as well and that for one product, reviews > might > > be important while for another related products might be more important. > > You need both and you need to remember the difference, especially before > > you know if these actions are important or distinct. > > > > You can cause these actions to be logged by using Javascript that > executes > > in the browser. Focus events or scroll events are both candidate methods > > for finding out more about what the user is doing. > > > > > > > > > > > > On Tue, Jan 14, 2014 at 7:47 AM, Manuel Blechschmidt < > > [email protected]> wrote: > > > > > Hi Abhishek, > > > > > > On 14.01.2014, at 16:24, abhishek kumar wrote: > > > > > > > Hi Ted, > > > > > > > > I'm new to mahout so I don't know how plugins for browser can be > > written > > > to > > > > incorporate mahout. Will you please explain in detail. > > > > > > Plugin for browsers have nothing to do with recommendations. Ted means > > > that you take a click log like the access log of the apache web server > > and > > > use this as the basis for a recommender system. > > > > > > Then you have to write a web application that serves the > recommendations > > > to the users. > > > > > > Here is a project that gives you all the details: > > > https://github.com/ManuelB/facebook-recommender-demo > > > > > > The project is based on Java EE technology. Java EE and Recommendations > > > are both very complex topics and you can easily spend 10 years with > both > > > and you won't discover all the details. > > > > > > I would recommend that you search for Mahout at youtube and slideshare > to > > > get some more details. > > > > > > > Thanks in advance. > > > > > > > > Regards > > > > > > > > > > > > On Mon, Jan 13, 2014 at 1:17 AM, Ted Dunning <[email protected]> > > > wrote: > > > > > > > >> On Sat, Jan 11, 2014 at 11:44 PM, Abhishek Kumar < > > > >> [email protected]> wrote: > > > >> > > > >>> For this I need to somehow integrate apache mahout to a browser. I > > also > > > >>> need > > > >>> to train my model on some server or database and then pipeline it > to > > > the > > > >>> client. > > > >>> > > > >>> Please help if you have any suggestions. > > > >>> > > > >> > > > >> Use log files. You can build a browser plugin that sends recent > URL's > > > to > > > >> an analytics server. > > > >> > > > >> If you are just working with friendly users, you might even just > use a > > > >> proxy server to get these logs. > > > >> > > > > > > -- > > > Manuel Blechschmidt > > > M.Sc. IT Systems Engineering > > > Dortustr. 57 > > > 14467 Potsdam > > > Mobil: 0173/6322621 > > > Twitter: http://twitter.com/Manuel_B > > > > > > > > >
